makepredictcall Utility Function for Safe Prediction

Description

A utility to help model.frame.default create the right matrices when predicting from models with terms like (univariate) poly or ns.

Usage

makepredictcall(var, call)

Arguments

var	A variable.
call	The term in the formula, as a call.

Details

This is a generic function with methods for poly, bs and ns: the default method handles scale. If model.frame.default encounters such a term when creating a model frame, it modifies the predvars attribute of the terms supplied by replacing the term with one which will work for predicting new data. For example makepredictcall.ns adds arguments for the knots and intercept.

To make use of this, have your model-fitting function return the terms attribute of the model frame, or copy the predvars attribute of the terms attribute of the model frame to your terms object.

To extend this, make sure the term creates variables with a class, and write a suitable method for that class.

Value

A replacement for call for the predvars attribute of the terms.

See Also

model.frame, poly, scale; bs and ns in package splines.

cars for an example of prediction from a polynomial fit.

Examples

require(graphics)

## using poly: this did not work in R < 1.5.0
fm <- lm(weight ~ poly(height, 2), data = women)
plot(women, xlab = "Height (in)", ylab = "Weight (lb)")
ht <- seq(57, 73, length.out = 200)
nD <- data.frame(height = ht)
pfm <- predict(fm, nD)
lines(ht, pfm)
pf2 <- predict(update(fm, ~ stats::poly(height, 2)), nD)
stopifnot(all.equal(pfm, pf2)) ## was off (rel.diff. 0.0766) in R <= 3.5.0

## see also example(cars)

## see bs and ns for spline examples.